A way to describe test traffic that has not been split evenly.

Instead, a smaller or greater proportion of users are allocated to either the control or the variant(s).

For example, in an A/B test with 1 control and 1 variant, unequal allocation occurs when 80% of traffic is directed to the control and 20% of traffic to the test variant (80/20).

Unequal allocation also occurs if there are multiple variants, but more traffic is allocated to the control than the variants, even if the amounts allocated to the variants is equal. For example, 40% directed to the control with 20% to each of the variants (40/20/20).

As described in this article, experimenters often believe unequal traffic allocation will yield accurate results faster. However, that's not the case.

Experimenters may also unequally allocate traffic to one version if they're not confident a variant will outperform. But this practice too, is not a reliable method of testing.

As a testing best practice, unequal allocation of traffic should be avoided, and all test traffic should be equally allocated.